Guest lecture · Dr. Shay Maymon · Deep Learning

A live business,
not a frozen dataset

Boss Software Solutions  ·  Lior Hermesh, Founder & CEO
→ next  ·  F fullscreen
The product

A live system — five screens

Not slogans. The real system, from daily use — watch a business become data.

The product · 1 / 5
Manager dashboard — sales vs last year, stock, tasks, churn alert, BI
Manager dashboard — sales vs last year, stock, tasks, churn alert, BI
The product · 2 / 5
Point of sale, mid-sale — basket, promo logic, live margin
Point of sale, mid-sale — basket, promo logic, live margin
The product · 3 / 5
Customer timeline — purchases, balance, calls, tasks over time
Customer timeline — purchases, balance, calls, tasks over time
The product · 4 / 5
Call → transcript + sentiment + emotion (a two-week demo)
Call → transcript + sentiment + emotion (a two-week demo)
The product · 5 / 5
Cloud add-on — Microsoft Fabric, Phase 1
Cloud add-on — Microsoft Fabric, Phase 1
Real-time

What “real-time” means here

The old way
  • Disconnected local terminal
  • Nightly sync
  • Manager sees yesterday
Our way
  • Native app on the registers
  • Talks to the hardware directly
  • Every action updates centrally — same split-second

Scan in branch A → stock drops everywhere now. No delay.

For DL: online inference while the customer is at the register.

Architecture

The Boss world

Internet Boss host multi-tenant Server farm Server 1 Server 2 Server 3 WebAPI server Native app POS · CRM · ERP Third-party clients Web stores Switchboard Delivery systems Phone app External CRM / ERP Facebook Leads …and more Native app — proprietary protocol Third-party clients — WebAPI
Data foundation

A modern data foundation

Gold Curated ML-ready pipelines Silver Cleaned · conformed Queryable across all businesses Bronze Raw · wire-faithful Noise · dupes · schema quirks Boss host all instances
Medallion lakehouse — raw → cleaned → ML-ready
The data

Four properties

Stream
Sequence models
Graph
Graph Neural Networks
Multimodal
Data fusion · text · voice · structured
Real-time
Low-latency inference

Real data is messy — Hebrew typos, schema drift, human noise. That’s the research.

Projects

Project menu

Applied DL
  • Next-basket & replenishment
  • Demand forecasting
  • Churn prediction
  • Anomaly & fraud detection
  • Price elasticity
Advanced DL
  • GNN for retail networks
  • Multimodal BI
  • Agentic AI systems
Full detail in the handout
Projects · deep dive

Churn prediction

1“Customers aren’t coming back”
2What is churn? No cancel button (non-contractual)
3Pet-shop 21-day vs fashion 3-month rhythm
4Model the Dynamic Inter-purchase Time
5Labels · features · strong baseline · metrics
→Catch them before they leave
What I provide

What I provide

Data

Tailored to your project — Bronze … Gold ML-ready. We won’t dump a raw file and vanish.

Compute

Server access & GPU time — not what your laptop can run.

Support

Relevant API access, docs, and technical guidance from me and the dev team.

research@boss-sws.com

Thank you

תודה רבה — אני פתוח לכל שאלה
research@boss-sws.com
1